xfer.model_handler.ModelHandler¶

class xfer.model_handler.ModelHandler(module, context_function=<function cpu>, num_devices=1, data_name='data')¶

Bases: object

Class for model manipulation and feature extraction.

Parameters:	module (`mx.module.Module`) – MXNet module to be manipulated. context_function (function) – MXNet context function. num_devices (int) – Number of devices to run process on. data_name (str) – Name of input layer of model.

Methods

`__init__`	Initialize self.
`add_layer_bottom`	Add layer to input of model.
`add_layer_top`	Add layer to output of model.
`drop_layer_bottom`	Remove layers from input of model.
`drop_layer_top`	Remove layers from output of model.
`get_layer_names_matching_type`	Return names of layers of specified type.
`get_layer_output`	Function to extract features from data iterator with model.
`get_layer_parameters`	Get list of layer parameters associated with the the layer names given.
`get_layer_type`	Return type of named layer.
`get_module`	Return MXNet Module using the model symbol and parameters.
`save_symbol`	Serialise model symbol graph.
`update_sym`	Update symbol attribute, layer names, and layer types dict and clean parameters.
`visualize_net`	Display computational graph of model.

Attributes

layer_names Get list of names of model layers.

drop_layer_top(num_layers_to_drop=1)¶

Remove layers from output of model.

Parameters:	n (int) – Number of layers to remove from model output.

drop_layer_bottom(num_layers_to_drop=1)¶

Remove layers from input of model.

Parameters:	n (int) – Number of layers to remove from model input.

add_layer_top(layer_list)¶

Add layer to output of model. model layers = (layer1, layer2, layer3), layer_list = [layerA, layerB] -> model layers = (layer1, layer2, layer3, layerA, layerB)

Parameters:	layer_list (list(`mx.symbol`)) – List of MxNet symbol layers to be added to model output.

add_layer_bottom(layer_list)¶

Add layer to input of model. model layers = (layer1, layer2, layer3), layer_list = [layerA, layerB] -> model layers = (layerA, layerB, layer1, layer2, layer3)

Parameters:	layer_list (list(`mx.symbol`)) – List of MxNet symbol layers to be added to model input.

get_module(iterator, fixed_layer_parameters=None, random_layer_parameters=None)¶

Return MXNet Module using the model symbol and parameters.

Parameters:	iterator (`mxnet.io.DataIter`) – MXNet iterator to be used with model. fixed_layer_parameters (list(str)) – List of layer parameters to keep fixed. random_layer_parameters (list(str)) – List of layer parameters to randomise.
Returns:	MXNet module
Return type:	`mx.module.Module`

get_layer_type(layer_name)¶

Return type of named layer.

Parameters:	name (str) – Name of layer being inspected.
Returns:	Layer type
Return type:	str

get_layer_names_matching_type(layer_type)¶

Return names of layers of specified type.

Parameters:	layer_type (str) – Return list of layers of this type.
Returns:	Names of layers with specified type
Return type:	list(str)

get_layer_output(data_iterator, layer_names)¶

Function to extract features from data iterator with model. Returns a dictionary of layer_name -> numpy array of features extracted (flattened).

Parameters:	data_iterator (`mxnet.io.DataIter`) – Iterator containing input data. layer_names (list(str)) – List of names of layers to extract features from.
Returns:	Ordered Dictionary of features ({layer_name: features}), list of labels Layer names in the ordered dictionary follow the same order as input list of layer_names
Return type:	OrderedDict[str, `numpy.array`], list(int)

get_layer_parameters(layer_names)¶

Get list of layer parameters associated with the the layer names given.

Parameters:	layer_names (list(str)) – List of layer names.
Returns:	List of layer parameters
Return type:	list(str)

visualize_net()¶: Display computational graph of model.

save_symbol(model_name)¶

Serialise model symbol graph.

Parameters:	model_name (str) – Prefix to file name (model_name-symbol.json).

layer_names¶

Get list of names of model layers.

Returns:	List of layer names
Return type:	list[str]

update_sym(new_symbol)¶

Update symbol attribute, layer names, and layer types dict and clean parameters.

Parameters:	new_symbol (`mx.symbol.Symbol`) – Symbol with which to update ModelHandler.